kreuzberg-dev/kreuzberg v4.0.0-rc.19

4.0.0-rc.19

on GitHub

[4.0.0-rc.19] - 2025-12-24

Added

• Font configuration API - Configurable font provider with custom directory support and automatic path expansion
• Homebrew bottle support - Pre-built macOS bottles for faster installation
• Environment variable configuration - KREUZBERG_MAX_REQUEST_BODY_BYTES and KREUZBERG_MAX_MULTIPART_FIELD_BYTES for API size limits
• Config file caching - Improved performance for TOML/YAML/JSON config file loading

Fixed

• Large file uploads (issue #248) - Files larger than 2MB are now accepted (configurable up to 100MB)
• Browser package Vite compatibility (issue #249) - Fixed missing pdfium.js in dist bundle
• Node.js missing binaries (issue #241) - Fixed resolution in Docker and pnpm monorepo environments
• Ruby gem native extension build - Simplified build system with proper linker path resolution
• Java E2E test compatibility - Regenerated tests for Java 25
• Docker ONNX Runtime - Pinned to version 1.23 for compatibility
• Font provider thread safety - Fixed race condition and graceful lock poisoning handling
• Custom font path validation - Added security hardening with symlink resolution and canonicalization

Changed

• BREAKING: Custom font provider now enabled by default (set enabled = false to disable)
• Default API size limit - Increased to 100MB (configurable via KREUZBERG_MAX_UPLOAD_SIZE_MB)
• TypeScript serialization - Replaced MessagePack + Base64 with direct JSON serialization

Performance

• 15-25% overall execution improvement, 30-45% memory reduction
• String allocation optimization (-2.57% CPU, -0.81% memory)
• Memory pool improvements (35-50% reduction: 60-135 MB → 30-80 MB)
• Post-processing pipeline optimizations (10-17% CPU improvement)
• TypeScript thread pool dynamically sized by CPU count

Removed

• Legacy code cleanup - Removed deprecated backward compatibility
  • TypeScript: KREUZBERG_LEGACY_SERIALIZATION environment variable
  • Go: 7 legacy error codes
  • Ruby: Ocr = OCR alias
  • Rust: Deprecated Metadata.date field
  • Cargo: 3 legacy feature aliases

Security

• Custom font directories validated with canonicalization
• Symlinks resolved to prevent path traversal attacks
• All custom paths validated before use